摘要
以长春市部分GF-2影像为实验数据,探讨了贝叶斯、k-最邻近、支持向量机、分类回归树和随机森林5种不同分类器的分类结果及精度。由于单一分类器的局限性,设计了一种多分类器的组合方法。考虑到Hellden值能够综合评定用户精度和制图精度,选取各地类具有最高Hellden值的分类算法,构建组合分类器,其最终的分类精度优于任何单一分类器的分类结果,达到优化分类器以提升分类结果的目的。
Based on part of GF-2 images of Changchun city, this paper first probes the classification results of five different classifiers: Bayes, k-nearest neighbor(KNN), support vector machine(SVM), Classification and regression trees(CART), and random forest. The comparisons of the overall accuracy and Kappa coefficient demonstrate that the descending order of these five classifiers in term of precision is as follows: SVM>Bayes>KNN>Random Forest>CART. Then this paper proposes a combined method of multi-classifiers according to the Hellden value. Research shows that the multi-classifier combination is a good method to improve the results of GF-2 image classification.
引文
[1] 韩启金,马灵玲,刘李,等.基于宽动态地面目标的高分二号卫星在轨定标与评价[J].光学学报,2015,35(7):372-379
[2] 孙攀,董玉森,陈伟涛,等.高分二号卫星影像融合及质量评价[J].国土资源遥感,2016,28(4):108-113
[3] 潘腾,关晖,贺玮,等.“高分二号”卫星遥感技术[J].航天返回与遥感,2015,36(4):16-24
[4] Du Peijun,Xia Junshi,Zhang Wei,et al.Multiple Classifier System for Remote Sensing Image Classification:A Review[J].Sensors,2012,12(4):4 764-4 792
[5] 宋英强,杨联安,许婧婷,等.基于Landsat-8卫星OLI遥感影像和AdaBoost算法的水体信息提取[J].测绘地理信息,2017,42(3):44-47
[6] 彭正林,毛先成,刘文毅,等.基于多分类器组合的遥感影像分类方法研究[J].国土资源遥感,2011(2):19-25
[7] 方文,李朝奎,梁继,等.多分类器组合的遥感影像分类方法[J].测绘科学,2016,41(10):120-125